Optimizer
Evolutionary search over NACA parameters, with each candidate scored by the surrogate instead of a CFD run.
Objective
02
Design space
0.0 – 7.0 %c
1.0 – 7.0 /10
8.0 – 18.0 %c
0.0 – 10.0 °
m/s
3194
Constraints
%c
518
Limit |Cm¼|
Stay inside training envelope
Algorithm & budget
12100
580
0999
Surrogate evaluations
1,240
Est. on GPU
≈ 6 min
Surrogate-in-the-loop optimization
Thousands of CFD-quality evaluations in minutes
A single RANS simulation of an airfoil takes minutes to hours. The surrogate predicts the full flow field in a fraction of a second, which makes population-based optimization practical.
1
Pick an objective
Efficiency, low drag at a lift target, maximum lift, robustness across AoA, or a Pareto trade-off.
2
Bound the design space
Choose which NACA parameters are free. Bounds are clipped to the surrogate's training envelope.
3
Add constraints
Minimum thickness for structure, a pitching-moment limit for trim.
4
Run
Each candidate is meshed by morphing and evaluated by the MeshGraphNet surrogate, streamed live.